Software Engineer, Accelerator Solutions & Technologies (phd)

Meta Meta · Big Tech · New York, NY

Software Engineer role focused on optimizing distributed AI/ML workloads and collective communications software for Meta's super-cluster AI/ML platforms. This involves contributing to developer infrastructure, simulation platforms, and understanding hardware acceleration techniques for ML inference and training.

What you'd actually do

  1. Contribute to our developer infrastructure, including simulation and HW emulation platforms, to enable performance measurement and optimization for Meta’s in-house accelerator programs
  2. Understand and contribute to the collective communications library, intended to be deployed on Meta’s AI/ML superclusters
  3. Support networking and compute hardware acceleration techniques to improve ML inference and training model performance
  4. Perform architectural analysis to ensure system designs meet performance, scalability, and reliability requirements
  5. Implement simulation models for Meta’s Accelerator ASICs, develop and analyze various scenarios to evaluate data center performance and identify potential improvements

Skills

Required

  • PhD degree in Computer Science, Computer Engineering, or relevant technical field
  • 2+ years experience in developing C++ codebase
  • 2+ years experience in developing Python codebase
  • Understanding of performance, benchmarking measurement, and optimization on collective communications and distributed at-scale model training
  • Understanding of the transport stack (e.g., RoCE) and its constraints particularly pertaining to interconnect and collective
  • Experience with SystemC
  • Knowledge of AI/HPC hardware requirements and specifications
  • Full-stack experience and understanding of AI/HPC systems, with a focus on the application layer and performance optimizations

Nice to have

  • prompt/context engineering
  • agent orchestration
  • responsible, ethical AI practices
  • PyTorch
  • CUDA
  • integrate AI tools to optimize/redesign workflows

What the JD emphasized

  • PhD degree
  • 2+ years experience in developing C++ codebase
  • 2+ years experience in developing Python codebase
  • Understanding of performance, benchmarking measurement, and optimization on collective communications and distributed at-scale model training

Other signals

  • AI/ML superclusters
  • ML inference and training model performance
  • AI/HPC hardware requirements